Teng and colleagues in WORK, October 13; AOTA evaluation guidance and Policy E.19
An AI-led occupational therapy intake can be easy to use and still leave people wanting a human conversation. In a small new trial, 25 of 30 adults preferred the human-led assessment.
In brief
- Each adult tried both formats, in randomized order.
- Usability was high for AI; compassion and satisfaction favored the human format.
- Experience ratings cannot establish clinical equivalence.
An intake tried two ways
Miini Minami Teng and colleagues published the study October 13 in WORK. The 30 adults completed a purpose-built remote AI assessment and a human-administered assessment one to two weeks apart; 28 also gave reflection interviews. The AI format scored 81.1 on a modified System Usability Scale. Perceived compassion was higher for the human format, with median ratings of 9.90 versus 7.53, and satisfaction also favored it.
Participants' agreement with the proposed goals was similar across formats. Strong agreement was more frequent after the human assessment on four of five goal items. With a sample this small, a finding of no statistically significant difference cannot establish equivalence. The reported results concern user experience; they do not establish diagnostic accuracy or improvement in everyday function.
What belongs in an OT evaluation
For a service considering AI intake, the distinction reaches beyond the questionnaire. AOTA's existing evaluation guidance describes two essential components: the occupational profile and analysis of occupational performance. The profile brings in the person's history, routines, interests and priorities. Performance analysis examines the abilities and barriers that shape participation. The therapist synthesizes that information to develop the intervention plan.
A completed intake form is therefore a starting point for further inquiry. A person may name cooking as a goal and later explain that the real difficulty is standing long enough to prepare a meal, remembering the sequence, or managing a crowded kitchen. Those examples are practical illustrations, separate from the trial's findings. They show why a useful first answer still needs room for a follow-up conversation.
Give a pilot a specific job
A clinic could test whether an approved AI tool helps people prepare for that conversation. Before a pilot, define what the therapist will review and how clients can correct a summary or request human help. Check whether the tool misses priorities that emerge later in the visit. Measure the time spent correcting its output alongside any time saved. These are suggested local checks, not outcomes demonstrated by this study.
AOTA's AI policy, effective since April 2025, retains professional judgment with the practitioner and calls for evidence-based review of outputs, accuracy and safety. This new study gives teams another reason to ask clients about the experience itself. A smooth interface may get someone through intake; the next task is finding out whether they felt understood.